• Vision for Robotics

    Vision is one of the most fundamental ways for robots to interact with the environments, and it can be used for many robot manipulation tasks such as scene understanding, object manipulation, and grasping planning. This project mainly addresses two problems in the field: Firstly, how to sense the target or the whole environment according to a specific task. Secondly, how to understand the sensing data which are more likely to be captured in an unstructured environment. Studying the problems will help the robot inspect the world more intelligently, and make crucial contributions to the deployment of robotic vision in industrial applications.

  • Deep Learning for Robotic Manipulation

    With the great success of deep learning in computer vision and computer graphics, many researchers are trying to transfer deep learning to the robotic manipulation field, to achieve task-specific smart decision and control. However, due to the fact that the manipulation system is totally different from deep learning applications in other fields in the aspects of data type, system modeling, etc., how to apply deep learning for robotic manipulation is a challenging problem. This project studies three kinds of deep learning methods on robotics: deep reinforcement learning, transfer learning, and few-shot learning. Through the researches on the three methods, we are able to achieve significant progress on real-world manipulation applications, such as object grasping, motion planning, and smart controlling.

  • 3D Measurement and Inspection

    3D measurement and inspection is one of the most fundamental issues in manufacturing industry. Recent advances in artificial intelligence and robot manipulation technology greatly facilitate the automatic 3D measurement. This project mainly addresses three problems in the field. Firstly,how to generate high quality scanning proposals. Second, how to produce full-coverage and efficient scanning paths. Thirdly, how to cooperate scanning scanner with mobile platforms to carry out efficient large-scale 3D measurement.

  • 3D Large-Scale Scanning and Data Processing

    Recent advances in Light Detection And Ranging (LiDAR) technology greatly facilitate the acquisition of 3D point data of largescale environments. The LiDAR data are coherent and inherently 3D, capable of faithfully representing real shapes of objects, benefiting a variety of applications such as digital city modeling, digital factory, 3d scene understanding and so on. However, the major drawback of LiDAR scanning is that the captured point data often suffer from low quality due to occlusion, motion, multiple reflections, etc., hindering its applications in many high level processing tasks. This project aims at designing efficient scanning strategy for different applications and processing the acquired large scale 3D data for the subsequent tasks.

  • Machine (Deep) Learning on 2D & 3D Data Analysis

    Deep learning based 2D & 3D Data Analysis has been attracting increasing attention of computer vision and graphics researchers recently. It is particularly relevant due to its importance for many applications such as self-driving cars, autonomous robots, virtual reality, and augmented reality. Behind the wide spectrum of applications lies the fundamental techniques in processing and analyzing 2D & 3D data, especially the popular learning-based ones.